Friday, November 7, 2014

How well do NTRP ratings predict winners at USTA League Nationals

I am closely following the 2014 USTA League 40 & over Men's 4.0 Nationals this weekend and wrote earlier in the day 1 review that using my Estimated Dynamic NTRP Ratings and taking the top-8 average for each team correctly predicted 14 of the 17 team matches today.  While the NTRP ratings goal is not necessarily to be predictive, that is a pretty good record.

But what about predicting the individual matches?  Here is what I found from today's matches.  The table below shows the ratings gap between the players/pairs and how often the favored team won vs upset.

GapFavoriteUpset
0.00-0.0546
0.05-0.151310
0.15-0.25177
0.25-0.35131
0.35-0.4562
0.45-0.5512
0.55-0.6520
0.65-0.7510
Total5728

Probably no big surprises here, the smaller the gap, the more likely it is that there might be an upset.  In fact, when the teams were nearly dead even, the lower rated player/pair won 60% of the time.  Now this is a small sample so don't read too much into it.

It is a little surprising to see there were more upsets when the gap was 0.35-0.55 than 0.25-0.35, but again, this is a small sample.

To compare, here is the same analysis for the 18 & over 4.0 Men played a few weeks ago.

GapFavoriteUpset
0.00-0.051814
0.05-0.152819
0.15-0.25246
0.25-0.35254
0.35-0.4583
0.45-0.5530
0.55-0.6510
0.65-0.7520
Total11946

In the 18 & over division, the higher rated player/pair seems to have won slightly more often.

What do you think?  Is this close to what you'd expect?